Fingerprinting problems and fingerprinting in VLC player

acoustid
audiofingerprinting
vlc-media-player
linux
Tags: #<Tag:0x00007f23c1b44448> #<Tag:0x00007f23c1b44240> #<Tag:0x00007f23c1b44010> #<Tag:0x00007f23c1e87e70>

#1

Hello to all musiclovers out there,

Here some remarks and a question from a (Musicbrainz/Picard) newbie,

I had quite some trouble getting the AcoustID / fingerprinting to work in Picard.

I installed Picard on a Linux Fedora 23 system.

In the help documentation I did not find any mentioning of the fact that you have to :

  • download and install the fingerprint calculator
  • get an AP key (what is that, I’m not a developer !)

Pressing the download button in the Picard options dialogue led me to the download page of AcoustID where I could see a number of “chromaprint…” files, but also the note:
“Most Linux distributions also have their own packages for Chromaprint.”

Aaah, very good, I started my Linux package manager (Yum/DNG) and looked for chromaprint, installed it and … and what now?

Eeuh… tell Picard where to find “it”, what is “it”, Various files (python scripts?) with “chromaprint” in the name, but which one to choose? None did work…

To make a long story short: Instead of installing chromaprint I downloaded the "fpcalc… " file from the forementioned page, unpacked it and, again… how/where to install??

So I just took a gamble and I copied it to home/.config/Musicbrainz directory, pointed Picard to it, got my API key(?) and HURRAY!! something worked.

That is to say: half of the files I threw at it came back with no results :frowning:

I had used VLC player before and ALL files showed up with the proper name/artists!

How is that possible?

VLC and Picard use the same system? Why doesn’t Picard show the same good matches that VLC could find?

To get better matches should I change some of the matching thresholds, and which?

For some tracks to give a (correct!) match, I had to change

Minimal similarity for file lookups: 1%
and
Minimal similarity for matching files to tracks: 1%

Seems rather odd to me… even 10% accuracy did not work for some files.

To resume
I think that Picard and the fingerprinting system is a fantastic piece of engineering, but a manual that also covers Linux installations and more details would be very welcome.

thanks and enjoy!


#2

fpcalc is the executable Picard needs. It is usually part of the chromaprint or some related package. As long as it is in your path Picard will autodetect it, so on most distributions this works out of the box since they install fpcalc as a dependency to Picard. I don’t know exactly about Fedora, but RPM fusion has a package: https://admin.rpmfusion.org/pkgdb/package/free/chromaprint-tools/

Picard could make it more clear what it needs here, but in this case it is IMHO also a packaging bug.

I don’t know what VLC is doing, in my VLC I could not even find any fingerprinting options. But one explanation is that VLC just uses plain AcoustID and the metadata provided there. When submitting fingerprints to AcoustID metadata can be provided, and AcoustID will return this. You can submit also MusicBrainz IDs, which will link an AcoustID to the MusicBrainz entry, but this is not required. So many AcoustIDs will provide some basic (not necessarily correct) metadata. Picard on the other side is supposed to tag against MusicBrainz data, an AcoustID fingerprint without a link to MusicBrainz is not used. According to the AcoustID statistics roghly 30% of AcoustIds are linked to a recording.

In the end it is hard to tell what is the cause in your case without specific examples.

The thresholds don’t effect the fingerprinting, they are for metadata comparison in different cases and the defaults work quite well in most cases. Actually you should most of the time not use the fingerprinting but the lookup function, and use fingerprinting for the difficult cases where you don’t have existing metadata available.


#3

I have ripped a few home-made CD’s that contain a selection of, for example, world music. Some of these tracks are definitely in the Musicbrainz database (I checked!). Ripping software appears to look at some kind of CD-id and uses that for referencing, which does not work, since it is a selection of tracks that I have put on a CD myself. So, in the end I have a collection of tracks without ANY data. I can type over all tags, but being lazy I use the fingerprinting option. Currently I have changed the thresholds to zero (0) in Picard to get best matches. Some of the tracks return without result, although, as mentioned they are in the database.

In VLC when you have tracks in your playlist you can right-click to get a menu with options. One of them is “Information…”, open this and there you will find a button “fingerprint”. This often gives a correct match in cases where Picard did not.